The present invention generally relates to power distribution networks and, in particular, a system and a method of efficiently detecting faults throughout the system using data analysis and probabilistic prediction techniques.
Power distribution networks often comprise many different nodes or points in the network and it is important to be able to detect when a node is powered and when a stretch of distribution line between nodes is faulted. Because the sensors used to detect the powered state at the many locations throughout the network may be noisy, it is preferable to view the state of the network at each point probabilistically. Furthermore, because of the scale of the network, computing the marginal probability density functions at each location in the graph is most often computationally impractical.